Sie sind auf Seite 1von 22

Journal of Public Economics 84 (2002) 91112

www.elsevier.com / locate / econbase

The optimal elasticity of taxable income


Joel Slemrod a , *, Wojciech Kopczuk b
a

The University of Michigan Business School, 701 Tappan Street, Ann Arbor, MI 48109, USA
b
University of Michigan, Department of Economics, Ann Arbor, MI 48109, USA

Received 7 February 2000; received in revised form 13 August 2000; accepted 7 December 2000

Abstract
The strength of the behavioral response to a tax rate change depends on the environment
individuals operate in, and may be manipulated by instruments controlled by the
government. We first derive a measure of the social benefit to affecting this elasticity. The
paper then examines this effect in the solution to the optimal income taxation problem when
such an instrument is available, first in a general model and then in an example when the
government chooses the income tax base. 2002 Elsevier Science B.V. All rights
reserved.
Keywords: Optimal income taxation; Tax avoidance; Tax administration
JEL classification: H21; H23; H26

1. Introduction
Behavioral elasticities are key to understanding the efficiency cost of non-lump
sum taxes. Standard treatments of optimal income taxation take as given
behavioral elasticities derived from immutable preferences, and characterize the
tradeoff between redistribution and the deadweight loss of progressive taxes.
Loosely speaking, larger elasticities imply that less progressive tax systems are
optimal. Most of these models have focused on the elasticity of labor supply, in
the belief that the relative price of leisure and goods is the most important margin
*Corresponding author. Tel.: 11-734-936-3914; fax: 11-734-763-4032.
E-mail address: jslemrod@umich.edu (J. Slemrod).
0047-2727 / 02 / $ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S0047-2727( 01 )00095-0

92

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

affected by taxes. Empirical evidence suggests that, at least for hours worked, the
aggregate (compensated) labor supply elasticity is quite small. However, increasing leisure is by no means the only possible margin of response to higher marginal
tax rates. When personal tax rates on ordinary income rise, evasion may increase,
businesses may shift to corporate form, there may be a rise in the consumption of
deductible activities such as charitable giving, and individuals may rearrange their
portfolios and compensation packages to receive more income as tax-preferred
capital gains. These responses to higher taxes, and all others, will show up in
declines in taxable income, and there is a growing body of evidence, that, at least
for high-income individuals, the elasticity of taxable income to the marginal tax
rate is substantial.
As Feldstein (1999) argues, under certain assumptions it is the compensated
elasticity of taxable income, regardless of whether its origin is responsiveness of
labor supply or some other behavior, that summarizes the efficiency cost of
taxation and therefore is the crucial parameter in models of optimal progressivity.
The idea is that all tax-induced behavior entails costs that will be incurred until, at
the margin, the private cost equals the tax saving. Certain qualifications to this
assertion are discussed in Slemrod (1998), including the importance of accounting
for shifts across tax bases (individual to corporate, for example) and across time
(e.g., a shift to deferred compensation or greater use of retirement accounts).
One critical qualification to the role of the elasticity of taxable income is that it
is a function not only of preferences, and is therefore not immutable. With regard
to labor supply, it is not too egregious to assume, as has become conventional in
the empirical literature, that its elasticity is a primitive value based on preferences.1
However, this assumption is very dubious when considering the other margins of
response to taxation. On the contrary, the magnitude of these behavioral responses
depends on a host of policy decisions about such things as the tax base and how it
is enforced. For example, the responsiveness of evasion depends on the enforcement regime, the ability to shift the form of business organization depends on the
rules governing the choice, the capital gains realization elasticity depends on how
carryover at death is treated, and the size of the tax shelter industry created by high
tax rates depends on the passive loss limitation provisions.2
Recognizing that the elasticity of taxable income is subject to policy control has
intriguing policy implications. For example, Slemrod (1994) demonstrates that the
1
It is more heroic, but standard, to assume that the elasticity is constant across individuals and time.
In addition, the labor supply elasticity may depend on labor market institutions such as work-week
regulations.
2
It is interesting to note the relation of our point to the influential critique of macroeconomic models
in Lucas (1976). As does Lucas, we raise the possibility that changes in policy affect behavioral
elasticities and, as a result, econometric procedures that ignore this possibility may give misleading
results. However, our main argument is different. We stress that affecting behavioral elasticities may be
the goal of the policy and not just its by-product, because it may make standard tax instruments (e.g.,
tax rates) more (or less) efficient.

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

93

optimal degree of income tax progressivity given a suboptimal setting of tax


enforcement can be below the globally optimal progressivity. The paper adopts the
metaphor of Okun (1975), in which redistribution is characterized as transferring
water from the rich to the poor in a leaky bucket, the degree of leakage
representing the efficiency cost of progressive taxes. Using Okuns language,
Slemrod (1994) raises the possibility that the leak can be fixed, albeit at some cost.
Moreover, there is an optimal rate of leakage that balances the cost of fixing the
leak against the improved efficiency of carrying water with a less leaky bucket. Put
less metaphorically, there are multiple instruments to an optimal tax system
problem, and an important attribute of a non-tax-rate, or administrative, instrument
is its effect on the elasticity of taxable income with respect to the tax rate. The
judicious choice of these instruments can be thought of as choosing the optimal
elasticity of taxable income. This paper examines what characterizes this optimal
elasticity.
The issue of optimal tax administration and enforcement, and their relationship
to the optimal rate structure, has been considered in the prior literature. Several
papers address this problem in a representative agent setting. Mayshar (1991)
provides general criteria for optimal administration and enforcement. Kaplow
(1990) and Cremer and Gahvari (1993) characterize the optimal enforcement and
optimal tax rate structure of commodity taxes. Also in regard to commodity
taxation, Yitzhaki (1979) and Wilson (1989) model the optimal tax base as a
tradeoff between the increased administrative costs, but smaller excess burden, of
a larger base. We draw heavily on their framework in Section 3 of this paper.
In a series of papers, Cremer and Gahvari (1994, 1996) examine the same issue
in the context of setting optimal income tax progressivity in a world of
heterogeneous individuals, and characterize optimal enforcement and the optimal
progressivity when tax evasion is present. Certainly, policy instruments that are
aimed at tax evasion, such as auditing, are examples of the non-tax-rate
instruments that this paper addresses. However, the specific framework Cremer
and Gahvari use in which evasion is a choice made by taxpayers under the
prospect of an uncertain audit makes it difficult to generalize their results to
instruments other than the intensity of enforcement and the penalties applied, and
focuses on the role of taxpayers risk preferences. None of the articles in either
strand of this literature frames the question, as we do in what follows, around the
elasticity of taxable income or characterizes the central problem as one of
choosing its optimal value.
We develop a framework that integrates and systematizes the results of these
literatures by identifying three types of effects that any policy instrument has: a
direct welfare cost, the revenue effect holding the strength of behavioral response
constant, and the effect on distortions caused by other tax instruments. Each
non-tax-rate instrument will feature each of these effects, but we stress that it is the
last effect that makes non-tax-rate policy variables fundamentally different from
tax rate instruments. In particular, it is this last effect that is responsible for

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

94

complementarities between tax design and tax administration, and thus its
understanding is key in the process of designing the optimal tax system. The
optimal setting of an administrative instrument must balance between its cost and
the benefit of reducing distortions; it is this trade-off that allows us interpret it as
choosing the optimal elasticity.
The paper proceeds as follows. First, we characterize the optimal linear income
tax solution when the government has an instrument which affects the elasticity of
taxable income. Recognizing that, in general, such an instrument has also a
redistributive impact as well as an effect on tax revenue, we isolate the value of
just affecting the elasticity. In a special case, when the elasticity is not a function
of the tax rate, the benefit of reducing it is directly related to the deadweight loss
of taxation. Next, we investigate at length a special case in which the policy
instrument which determines the taxable income elasticity is the broadness of the
tax base. In this context, we demonstrate some propositions about the optimal tax
system and the optimal elasticity of taxable income, in particular how they depend
on the primitive parameters of the problem such as the social welfare function and
the cost of administration.

2. The optimal linear income tax with a policy instrument that affects the
elasticity of response
We first consider the general problem of choosing the optimal linear income tax
in a multi-person economy. In this problem the government must raise R in
revenue.3 In the standard version of this problem, it can choose two parameters of
a linear income tax, the marginal tax rate (t) and a uniform lump-sum tax (T ).
Raising all of the required revenue using the lump-sum tax would minimize excess
burden, but the population is comprised of people with a distribution of abilities to
earn income, and the government objective function is such that it would be
willing to accept some inefficiency in exchange for less inequality in the
distribution of welfare. We assume that the society consists of a large number of
individuals indexed by the parameter y, a measure of ability, with the distribution
function F( y) normalized as e F( y) dy 5 1.
We modify the standard problem by assuming that the government controls an
instrument m, which inter alia reduces the elasticity of taxable income with respect
to the tax rate, but has a cost of implementation of A(m), A9(m) . 0. Given the
parameters of the tax system, each individual maximizes the standard utility
function that gives rise to the indirect utility function v(t, T, m, y) and the taxable
income function I(t, T, m, y).4 The total tax paid by the individual is equal to
3

We do not inquire about, nor do we model, how the government disposes of R.


Because we allow for primitives other than the utility function (for example, an avoidance
technology), it may be impossible to recover taxable income from the indirect utility function, so that
they may be independent.
4

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

95

tI 1 T. To keep our framework general, we do not specify the utility function, nor
do we define taxable income explicitly. Some examples of problems that fit into
this framework are:
A standard model with a consumption (C)labor (L) choice and an instrument
m that affects individuals preferences between these two goods (e.g., a public
good or education of taxpayers about services financed through taxes). In this
case, the budget constraint is C 5 yL 2 T 2 tI. Maximization of the utility
function u(C, L, m) gives rise to the indirect utility function and taxable income
defined by I 5 yL.
A model (as in Mayshar (1991) or Slemrod (2001)) with tax sheltering S, that
costs M(S, yL, m). The utility function u(C, L) is maximized subject to the
budget constraint C 5 yL 2 T 2 t( yL 2 S) 2 M(S, yL, m); in this case taxable
income is defined as I 5 yL 2 S.
Our example of Section 3. An individual consumes a large number (continuum)
of commodities, described by a bundle hCi j i [[0,1] with a utility function u(C).
Labor supply is present in the model, but it is inelastic. The policy instrument
m [ [0, 1] determines the size of the commodity tax base (equivalently, goods
i [ [m, 1] are income-tax-deductible), and the budget constraint takes the form
of e01 Ci di 5 y 2 T 2 t( y 2 em1 Ci di). Consequently, I 5 y 2 em1 Ci di.
The government is assumed to choose its instruments to maximize a generalized
utilitarian social welfare function v, so its problem can be expressed as
max
t,T,m

E v(v(t, T, m, y)) dF( y),

(1)

subject to the budget constraint


T1

E tI(t, T, m, y) dF( y) 5 R 1 A(m).

(2)

We use the concept of the compensated (Hicksian) taxable income denoted by I c (t,
T, m, y, u) (where u is the utility level), which is the taxable income implied by a
solution to the consumers expenditure minimization problem.5 Compensated
demand is a well-established concept, and the compensated taxable income is an

5
In the expenditure minimization problem, the consumer minimizes spending, subject to the
minimum utility constraint. Consequently, we assume that the dual approach is valid, so that the
identity I(t, T, m, y; E(t, T, m, y, u)) 5 I c (t, T, m, y; u), where E( ? ) is the expenditure function, holds.
Differentiating this identity with respect to the tax rate t, and observing that E( ? ) / t 5 I, leads to the
Slutsky equation in the text. Notation I and I c distinguishes uncompensated taxable income from
compensated taxable income, the difference being that one is a function of exogenous income while the
other is a function of the utility level.

96

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

aggregation of compensated demands for relevant commodities. Analogously, the


Slutsky-type equation for taxable income can be derived,
I c
I
I
]]] 5 ]]] 2 I ] ,
E
(1 2 t) (1 2 t)

(3)

where I / E is the income effect. Denoting the Lagrange multiplier by l, the


three first-order conditions, rearranged using the standard results of consumer
optimization (Roys identity and the Slutsky equation), are as follows
I
DD dF( y),
E S 2 v 9m 1 lS1 2 t ]
E
I
t
t: 0 5 E S 2 Iv 9 m 1 lIS1 2 t ] 2 ]] DD dF( y),
E 1 2 t
I
m: 0 5 E Sv 9v 1 lSt ] 2 A9(m)DD dF( y),
m
T: 0 5

(4)
(5)
(6)

where m denotes the marginal utility of exogenous income (the Lagrange


multiplier from the consumers maximization problem), and ; (1 2 t /I c)(I c /
(1 2 t)) is the compensated elasticity of taxable income with respect to the
net-of-tax rate (1 2 t). Stating the elasticity with respect to (1 2 t), rather than with
respect to t, is standard in this literature.
For given m, conditions (4), (5) and the budget constraint determine the solution
to the standard optimal taxation problem. For example, Atkinson and Stiglitz
(1980) combine them to form
t
cov(I, g)
]] 5 2 ]]],
12t
I dF( y)

(7)

where g ; (v 9 m /l 1 t)(I / E) is the marginal social valuation of a lump-sum


transfer to an individual. This expression implicitly determines the optimal tax rate
and suggests that optimal progressivity is inversely related to the compensated
elasticity of taxable income which, in the AtkinsonStiglitz model, is equivalent to
the compensated elasticity of labor supply.
In the model presented here, the critical elasticity may depend on the choice of
m, so that the optimal progressivity of the tax system cannot be determined
independently of m. To pursue this idea further, we can write the first-order
condition for m, expression (6), as
I
E S]vm g 1 t ]
dF( y) 5 A9(m),
m D
c

(8)

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

97

where I c / m refers to the compensated response of taxable income.6 It is


insightful to further decompose the impact of m on compensated taxable income.7
To do that, notice that, for any t * , t * I c (t * , T, m, y; u) 5 e0t * st(I c / t) 1 I cd dt 5 e0t *
I c (1 2 t / 1 2 t) dt. Increasing the tax rate by t increases revenue by I c t, but it also
stimulates a substitution response that reduces taxable income by (t / 1 2 t) percent
for each percent increase in the tax rate. Differentiating this formula with respect
to m, we can write
t*

E S

t*

D E

I c
I c
t
tI c
t ] 5 ] 1 2 ]] dt 2 ]] ] dt.
m
m
12t
1 2 t m
*

(9)

The first term on the right-hand side of this expression represents the direct effect
on tax revenue of changing m. The second term represents the effect on revenue of
decreasing the compensated elasticity of taxable income. The impact of this
change is positively related to the tax rate.
First-order condition (8) can now be written as

E1

t*

E S

t*

D E

vm
I c
t
tI c
]
g 1 ] 1 2 ]] dt 2 ]] ] dt dF( y) 5 A9(m).
m
m
12t
1 2 t m
0

(10)

Expression (10) says that, at the optimum, the benefit of increasing m should equal
the marginal administrative cost. The benefit is the sum of the direct effect on
social welfare (denominated in dollars), the revenue effect holding the elasticity
constant, and the benefit of changing the elasticity. Note that all expressions are
evaluated at the constant utility level (in other words, integration is along the
indifference curve).
The last term of the left-hand side of expression (10), 2 e e0t * (tI c / 1 2 t)( /
m) dt dF( y), is of particular note. It implies that the value to the social planner of
a policy instrument is higher (holding its other effects constant), the more an
instrument decreases the compensated elasticity of taxable income. Furthermore,
the benefit of lowering the elasticity increases with the tax rate and with taxable
income. Thus, the more egalitarian is the social objective function (which
manifests itself in higher progressivity), the greater is the benefit from using this
instrument, because the reduction of the compensated taxable income elasticity is
more valuable. This suggests that the elasticity at the optimum, the optimal
elasticity, should be lower in more egalitarian societies. We return to this issue in
Section 3.
6

In deriving expression (8), we make use of the Slutsky-type equation for the effect of m:
(I / m) 5 (I c / m) 1 (I / E)(vm /m ).
7
Although the decomposition that we consider concerns compensated quantities, a similar approach
may be applied to quantify the benefit of a change in the uncompensated elasticity.

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

98

An interesting special case occurs when the elasticity does not depend on the
marginal tax rate.8 We can then show that e0t * (tI c / 1 2 t) dt 5 21 (e0t * I c dt 2 t * I c).
Interpreting I c as the demand for taxable commodities and t as their price, the term
in brackets can be thought of as the standard (i.e., not including A(m)) deadweight
loss from taxation. It is (minus) the difference between the amount collected in
distortionary taxes and the amount that a discriminating (i.e., one that can impose
a person-specific, arbitrary non-linear marginal tax schedule) tax authority could
collect, with demand evaluated at the constant utility level. Then the benefit of
reducing the taxable income elasticity (the third term of the left-hand side of Eq.
(10)) is
t*

EE

tI c
]] ] dt dF( y) 5 m
1 2 t m

t*

E 1E I dt 2 t I 2 dF( y),
c

* c

(11)

where m 5 2 ( / m) / is the proportionate change in elasticity resulting from


an increase in m, defined to be a positive number. Eq. (11) says that the value to
the social planner of reducing the compensated taxable income elasticity is
proportional to the excess burden.
Although this approach provides a clear intuition for the benefit of a change in
the elasticity, it may be difficult to calculate in practice, because all terms are
evaluated at a constant utility level. In Appendix A, an alternative expression is
derived with the benefit of a change in elasticity evaluated at a constant income
level. The benefit of a change in elasticity can be shown to have a similar form as
before:
t*

tI
] dt dF( y),
EE ]]
1 2 t m

(12)

but the integration takes place along the budget line instead of an indifference
curve.
The model presented in this section isolates in a very general setting the role of
elasticity control in an optimal tax system. However, the generality of the
instrument m makes it difficult to establish insightful propositions about the
expanded optimal progressivity problem, although it suggests likely relationships
between the optimal taxable income elasticity and the governments goals.
Practical examples of m range from the extent of audit coverage, to the definition
of a pass-through entity, to broadening the tax base. In addition to its effect on the
elasticity of taxable income, any instrument has both efficiency and distributional
8
Even if this is not literally true, it may be a useful approximation if the effect of m on elasticity is
much greater than the effect of t. It may also be a useful benchmark for empirical applications, where it
is conventional to assume that elasticities do not depend on prices.

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

99

impacts. To gain further insight, we next examine the optimal elasticity question in
a particular setting in which the elasticity is determined by how broadly the tax
base is defined.

3. A model of optimal progressivity and the optimal tax base


The structure of the problem is similar to the general one described before. The
government must raise revenue of R, and can choose three instruments: a
lump-sum tax T, a marginal tax rate t, and a third instrument that, inter alia, affects
the elasticity of taxable income. Rather than being very general about this third
instrument as we were in the previous section, we are now quite specific: it is the
breadth of the tax base. More details follow.

3.1. The consumer s problem


Consider an economy in which there is a continuous number of commodities.
Commodities are indexed by a number i [[0, 1], and they are denoted by Ci . The
bundle of commodities is written as C (C:[0, 1] R 1 ). In other words, a bundle of
consumption goods is a vector containing a continuous number of elements.9
Each individual is characterized by a wage rate of y, and provides inelastically
one unit of labor. She maximizes a CobbDouglas utility function given by
u(C) 5 exp(e01 hi ln Ci di), where hi is the weight of good i.10 Weights are
normalized so that e01 hi di 5 1.
Pre-tax prices of all commodities are set equal to one. The consumer faces a tax
system characterized by three parameters: t, T and m, where m, 0 # m # 1, denotes
the size of the tax base. In a consumption tax framework, this implies that
commodity i is taxed as long as i , m. Equivalently, in an income tax framework
it implies that spending on commodity i [ [m, 1] is deductible from taxable

9
We assume that the consumption space includes only measurable functions, and we employ
Lebesgue integrals in what follows.
10
The setup of our model is closely related to the models of Yitzhaki (1979) and Wilson (1989).
Wilson (1989) considered a more general CES function. In an earlier version of this paper we
generalize the results that follow to the CES case. Most of the propositions in this paper go through, but
some of the proofs are significantly more complicated. The main source of complexity is that, in the
CES case, the compensated elasticity of taxable income is a function of the tax rate. One result that is
affected is our Corollary 1, which must be qualified to state that more egalitarian societies have a lower
optimal elasticity of taxable income to the extent that the elasticity of substitution is less than or equal
to one. When it is greater than one, it is still true that the tax base increases with egalitarianism (acting
to reduce the elasticity), but an adjustment in the tax rate may affect the elasticity in the opposite
direction. In the earlier version, we present an example in which the net effect is an increase in the
elasticity.

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

100

income. In a single-period framework like this, these two interpretations are


equivalent.
For future reference, following Wilson (1989) and Yitzhaki (1979), we
introduce the notation g (m) ; e0m hi di, where g (m) is the expenditure share
corresponding to a tax base of m. Note that g (m) is a monotone transformation of
m, and as such it may equivalently be used as a policy instrument. The parameter t
is the marginal tax rate imposed on taxable income; we will also use the notation
t ; 1 2 t to denote the net-of-tax rate. The parameter T is the lump-sum tax. To
facilitate the analysis, with no loss of generality, we use the convention that the
lump-sum tax is deductible from taxable income, so that taxable income is equal to
I 5 y 2 em1 Ci di 2 T. Consequently, the budget constraint of an individual is
1

E C di 5 y 2 T 2 t( y 2E C di 2 T ).
i

(13)

The utility function is CobbDouglas, so that the solution to the consumers


problem is

Ci 5

(1 2 t)hi ( y 2 T ) i , m

hi ( y 2 T )

,
i $m

(14)
1

and thus taxable income is equal to I 5 y 2 T 2 em hi ( y 2 T ) di 5 g ( y 2 T ).


Substituting the optimal choices of Ci into the utility function we can also derive
the indirect utility function: v(t, g, y 2 T ) 5 a ( y 2 T )t g, where a is a constant.
Henceforth, we normalize a to be one.
Finally and critically, but demonstrated in Appendix B, the compensated
elasticity of taxable income with respect to the net-of-tax rate, denoted , is simply
1 2 g. Thus, with these assumptions, the choice of tax base immediately
determines the central elasticity in the model. Intuitively, the larger is the set of
untaxed commodities to which the consumer can turn to in the face of taxation, the
larger is the elasticity.11

3.2. The government s problem


The governments problem is to choose t, T, and g to maximize the social
welfare function, subject to its budget constraint. To be precise, its problem is
max

E w(v(t, g, y, T )) dF( y)

(15)

subject to
This is also true in the CES case, where this elasticity is equal to 5 s (1 2 g /g 1 (1 2 g )t 12 s ), s
being the elasticity of substitution. This expression is decreasing in g.
11

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

T 1 tg (Y 2 T ) 5 A(g ) 1 R ,

101

(16)

where Y 5 e y dF( y). The choice of the tax base determines the administrative cost
A(g ). Recall that all individuals maximize a homogeneous utility function, and
following Wilson (1989, p. 1199) we assume that, for given g, those commodities
which minimize A are included in the tax base, so that A9(g ) . 0 and A0(g ) . 0.12
Maximization of (15) subject to (16) leads to the following first-order
conditions with respect to T, t and g, respectively,

E [2w9t 1 ls1 2 tgd] dF( y) 5 0,


E [2w9t g( y 2 T ) 1 lg( y 2 T )] dF( y) 5 0, and
E [t ln(t)w9( y 2 T ) 1 lt( y 2 T )] dF( y) 5 lA9(g ).
g

(17)

g 21

(18)
(19)

We will denote by X 5 tg e ( y 2 T ) dF( y) 5 tg (Y 2 T ) the revenue from


distortionary taxes, so that the budget constraint becomes T 1 X 5 A(g ) 1 R.
Given the revenue requirement of R, if X and g are known, the value of T can be
easily found. For this reason any point in (g,X) space represents a unique tax
system satisfying the budget constraint. We will concentrate on X as a measure of
progressivity, because in this model progressivity depends not only on t, but on g
as well.
The standard, or what we will call the Mirrlees, optimal income taxation
problem can be thought of as solving for the optimal t and T, for a given tax base
of g. Its solution is characterized by the first-order conditions (17) and (18), plus
the budget constraint. Solving this problem for all values of g defines a curve in
(g, X) space that we will refer to as a Mirrlees curve, which is depicted in Fig. 1.
It determines the optimal progressivity for a given tax base or, equivalently, for a
given compensated elasticity of taxable income.
Contrast this to the problem considered by Yitzhaki (1979) and Wilson (1989),
who take the revenue requirement and the lump-sum portion of the tax system (T )
as given, and solve for the optimal g and t. The solution to this problem is
characterized by first-order conditions (18) and (19), plus the budget constraint.
12
The assumption A9(g ) . 0 was introduced by Yitzhaki (1979), who provided a simple justification
for this shape. We make this assumption for convenience, but it is not required for our results. In
particular we know that, as long as a global optimum exists, administrative costs must be increasing in
its neighborhood otherwise one could extend the tax base and decrease both distortions and
administrative costs. Consequently, without this assumption, our propositions should be read as
describing the situation in the neighborhood of the optimum only.
We recognize, but do not address, the possibility that, depending how the tax system is administered,
A(1) could be less than A(g ), g , 1. In other words, a completely broad base may be less costly to
administer than one with exclusions.

102

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

Fig. 1. The Mirrlees and YitzhakiWilson curves and the optimal tax system.

Solving this problem for all values of T defines a curve in (g,X) space that we will
refer to as a YitzhakiWilson (Y-W) curve, also shown in Fig. 1. It determines the
optimal tax base for a given degree of progressivity.
Note that using (18) to get an expression for l, and substituting for l in (19)
yields
A9(g ) 5st 1 t ln td(Y 2 T ).

(20)

This expression is almost identical to Yitzhakis Eq. (5), and is a special case of
Wilsons Eq. (18), that characterize the optimal tax base. The only difference is
that the term Y 2 T replaces y in order to account for multiple individuals and the
lump-sum tax. Together with the budget constraint, expression (20) characterizes
the optimal choice of g and X (or t), given T. The right-hand side of this
expression is the marginal social benefit of expanding the tax base, which is equal
to the marginal reduction in the excess burden from raising a given amount of
revenue with a larger tax base. The left-hand side is simply the marginal resource
cost of expanding the tax base. In the models of Yitzhaki and Wilson, the amount
of revenue to be raised by distortionary taxes is fixed, so that the only problem is
whether to use a narrow-base, high-rate system, or a broader-base, lower-rate
system. In the problem considered here, though, the government can also reduce
distortionary taxes (and, therefore, redistribution) by making greater use of lumpsum taxes. Because with a CobbDouglas (or, more generally, a CES) utility

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

103

function the income elasticity of all goods is exactly one, there are no distributional implications to whether a given amount of distortionary tax is raised via
a narrow or broad base; thus, for a given R the value of X is a sufficient statistic
for evaluating the tax systems progressivity, as is the value of T.
Eq. (20) can be decomposed as in Section 2. Recall that the social benefit of a
reduction in elasticity is given by 2 e e0t * (tI / 1 2 t)( / g ) dt dF( y). Substituting
expressions for I and , and integrating, this benefit becomes g (Y 2 T ) e0t *
(t / 1 2 t) dt 5 2 g (ln(t ) 1 t)(Y 2 T ). It is straightforward to show that this
expression is positive and increasing in both g and t. Consequently, without this
effect the benefit of an increase in g (the right-hand side of expression (20)) would
be smaller, and the incentive to extend the tax base to meet the egalitarian goals
would be weaker.
How important is the role of a change in the elasticity? To quantify it, one can
calculate the share of this benefit in the total benefit of a change in g, the
right-hand side of Eq. (20),
2 g (t 1 ln(t ))(Y 2 T )
]]]]]] 5 g B(t),
(t 1 t ln(t ))(Y 2 T )

(21)

where B(t) 5 (t ln(1 2 t) /t 1 ln(1 2 t) 2 1)21 . One can show that B / t . 0,


B(0) 5 1, and lim t 1 B(t) 5 1 `. Consequently, for small values of t, the share of
the benefit due to the change in the elasticity is approximately g. For sufficiently
large t (but less than one), we have that g B(t) . 1, meaning that the benefit of a
change in the elasticity exceeds the total benefit, i.e., that the sum of all other
effects is a net loss. Consequently, it is only the change in the elasticity that makes
an increase in g worthwhile (note that the total benefit, the right-hand side of (20),
is always positive). This should not be too surprising: if the elasticity does not
change, there would be limits to redistribution so that, at some point, increments to
progressivity achieved by increasing g would reduce welfare.
We are now ready to show the relationship between the social welfare function
and the optimal tax system.

3.3. Results
In the problem we allow all of the policy variables, t, T, and g, to be
endogenous. Thus, the global optimum lies at the intersection of Mirrlees and
YitzhakiWilson curves, determining both the progressivity of the tax system and
the tax base (which determines the elasticity).
We seek to understand how the optimal tax system depends on the key aspects
of the environment. The first of our propositions depends on the following lemma,
which is demonstrated in Appendix B.
Lemma 1. The YitzhakiWilson curve is upward sloping.

104

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

We can now state our first proposition (proven in Appendix B) as


Proposition 1. A more egalitarian society will feature a larger tax base (and a
more progressive tax system).
Notice that the Y-W curve does not depend on the social welfare function: it is
defined by the budget constraint and Eq. (20), and both of these are invariant to
changes in the social welfare function.13 Consequently, an increase in progressivity
corresponds to a shift along the Y-W curve. Thus, progressivity and the tax base
either both increase or they both decrease.
It is not surprising that more egalitarian societies have a more progressive tax
structure. However, in this model, the social planner can increase progressivity by
either increasing the marginal tax rate or by increasing the tax base. Both policy
changes affect tax revenue in a similar way, but increasing the tax base reduces
distortions (and increases administrative costs). Our previous analysis demonstrates
that extending the tax base is more than just a complementary way of making
taxes more progressive. By reducing the strength of the behavioral response, an
increase in the tax base makes a higher marginal tax rate more effective in
achieving egalitarian goals. Consequently, without this effect, the benefit of an
increase in the tax base (the right-hand side of Eq. (20)) would be smaller, and it is
not clear that the tax base would have to increase at all.
Corollary 1. More egalitarian societies have a lower compensated elasticity of
taxable income.
Proof. The compensated elasticity of taxable income is equal to 1 2 g, and
Proposition 1 established that g is higher in more egalitarian societies. h
Our second proposition depends on the three lemmas which also underlie the
pattern of curves in Fig. 1.
Lemma 2. An increase in egalitarianism shifts the Mirrlees curve up.
Proof. In Appendix B. h
Lemma 3. Near the intersection of the two curves corresponding to the global
optimum, the YitzhakiWilson curve has a larger slope.
Proof. By Lemma 1 the Y-W curve is upward sloping. A change in the welfare
function does not affect the Y-W curve, because neither Eq. (20) nor the budget
13

A change in the social welfare function does not affect the tradeoff between t and g. Because of the
homogeneity of the utility function, neither of these instruments has distributional implications.

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

105

constraint are affected. By Lemma 2, an increase in the concavity of the welfare


function shifts the Mirrlees curve up, and by Proposition 1 the outcome of this is
an increase in g and X. This is only possible if the Y-W curve is steeper than the
Mirrlees curve. h
This lemma corresponds to a diagrammatic description of the experiment in
Proposition 1. An increase in egalitarianism shifts the Mirrlees curve up and, given
the way the curves intersect, it leads to higher progressivity and a broader tax base.
Lemma 4. Assume that the social welfare function is isoelastic. Then, the Mirrlees
curve is upward sloping if A9(g ) is small enough.
Proof. In Appendix B. h
Proposition 2. Assuming an isoelastic welfare function and small enough
administrative costs, an increase in the cost of maintaining a broad tax base (i.e.,
A9(g *) increasing while A(g *) stays constant) will decrease optimal progressivity
(as well as the tax base).
This proposition is demonstrated in the appendix. It is the complement to
Proposition 1. In this case, changing the marginal cost of administration does not
alter the solution to the Mirrlees problem, so that the optimum (g *, X * ) lies on the
original Mirrlees curve. The changing administrative cost does, though, shift the
Y-W curve, i.e., it changes the optimal mix of t and g to raise a given amount of
distortionary tax. In fact, it must shift upward, so that for a given g, the optimal X
is higher. Thus, the new optimum features a smaller tax base, which is not
surprising, as well as a less progressive tax system, because of the higher cost of
collecting those revenues.
Corollary 2. A decrease in the cost of maintaining a broad tax base leads to a
lower elasticity of taxable income.
This corollary is an immediate consequence of Proposition 2, because the elasticity
of taxable income moves inversely with the tax base.
In a sense, Corollary 2 is parallel to Corollary 1: in the latter the benefit of
extending the tax base increases, and in Corollary 2 the cost decreases. Both
situations naturally lead to a broader tax base being optimal. However, Corollary 1
is more fundamental, because when egalitarianism increases a change in the
compensated elasticity is a crucial part of the response of the social planner.
Consequently, if this effect were not present, the optimal tax base would not have
to increase. In the present case, as long as the optimum is interior (as we assume
throughout the paper), the instrument that becomes cheaper will be used more,

106

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

regardless of its effect on the elasticity (although exactly how much more it will be
used depends on this effect).
Fig. 1 and the model that underlies it may also be used to demonstrate the
danger of designing tax policy without regard for whether the existing elasticity of
taxable income is appropriate. Consider an initial tax system corresponding to
point A. Based on the elasticity of taxable income for that tax base, the
government may (correctly) decide to choose the tax system represented by point
D, that is, to reduce progressivity. However, the global optimum O corresponds to
a higher level of progressivity than point A.14 Even if one recognizes that other
instruments are available, but fails to recognize their effect on the taxable income
elasticity, the results may still be misleading. The perceived Mirrlees curve would
be flatter, but still crossing point D, because the elasticity would be assumed to be
unaffected by g, and thus the perceived cost of progressivity at higher levels of g
would be higher than it is in reality. The perceived YitzhakiWilson curve would
be to the left of the actual one, because the benefit of a change in g would be
underestimated. The perceived global optimum would correspond to lower
progressivity and a lower tax base than the actual one, and the corresponding level
of progressivity could be lower than at A.
The choice of the income or commodity tax base is an illustration of how
policymakers have some degree of control over the elasticity of taxable income to
the tax rate. While the details of other examples may certainly differ from what we
have presented here, the underlying intuition will remain.

4. Implications
The optimal progressivity of the tax system depends inversely on the compensated elasticity of taxable income with respect to the net-of-tax rate. That elasticity
is not entirely an immutable function of consumer preferences and production
technologies, but is subject to manipulation by government policy with respect to
tax administration and enforcement as well as the choice of tax base.
The notion that the elasticity of taxable income can be chosen via the setting of
certain policy instruments has noteworthy implications both for policy and the
empirical analysis of taxation. Although Section 3 focused on cross-jurisdictional
implications, it is of interest to speculate about its implications for a single
countrys tax policy over time. For example, consider its implications for the last
major reform in the United States, the Tax Reform Act of 1986 (TRA). TRA
lowered statutory tax rates and at the same time broadened the tax base and
restricted the use of tax shelters, both of which arguably reduced the elasticity of
taxable income. This is consistent with the fact that empirical estimates of taxable
14

For completeness, observe that it is also possible that the level of progressivity suggested by the
standard analysis may be too high, for example if one starts at a point such as B.

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

107

income elasticities based on the behavioral response to TRA (Auten and Carroll,
1999; Feldstein, 1995; Moffitt and Wilhelm, 2000) are larger than those based on
the behavioral response to subsequent (1990 and 1993) tax rate changes (Carroll,
1998). Although differences in methodology and offsetting biases due to unobserved trends may be partly responsible for the decrease, this paper raises the
possibility that the base broadening provisions of TRA reduced the true taxable
income elasticity between the two periods. More troublingly, this paper raises
doubt about longitudinal analyses of tax reforms which feature changes in both tax
rates and (in the case of TRA, as a policy objective) changes in the elasticity of
taxable income. Comparing pre- and post-reform behavior to identify the elasticity
is problematic, to say the least, because the elasticity after the reform is different
from the elasticity before the reform.
If one takes seriously the idea that actual tax policy decisions reflect a solution
to a maximization problem like that posed here, many intriguing predictions arise.
For example, one could argue that, once the Tax Reform Act of 1986 was passed
and policy makers became aware that the taxable income elasticity (and the
marginal social cost of increasing progressivity) had declined, they appropriately
increased the progressivity of the tax system by raising the top income tax rates in
1990 and 1993.15 We are aware that many factors are at work in producing tax
legislation, and we are certainly not suggesting that this factor dominates the
others, but we are raising the possibility that a policy maker who controls the
elasticity can affect the resolution of the progressivity issue. The principal message
to policy makers is to be wary of decisions about progressivity based on the
existing elasticity of taxable income: this elasticity is itself subject to control, and
may be suboptimal. Progressivity, and all of the other aspects of a tax system,
must be jointly optimized.
Another natural research agenda is to test the cross-jurisdictional implications of
this model. For example, exogenous measures of egalitarian social preferences
should be positively correlated not only with effective tax progressivity, but also
the breadth of the tax base, and be negatively correlated with the observed
elasticity of taxable income. The simultaneous presence of multiple taxes such
as income tax and value added tax means that the underlying normative model
will have to be extended in straightforward ways.

Acknowledgements
We are grateful to the referees, Julie Cullen, Bev Dahlby, Austan Goolsbee, Jim
Hines, and to members of the University of Michigan Public Finance bag lunch
workshop, the National Bureau of Economic Research Public Economics work15

In terms of Fig. 1, TRA moved the income tax system from D to E, and the top rate increases of
1990 and 1993 moved it from E to O.

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

108

shop, the Harvard Public Finance seminar and the Stanford Institute for Theoretical Economics workshop for comments on an earlier draft.

Appendix A. Proofs related to Section 2


Alternative expression for the benefit of a change in the elasticity: write the
first-order condition (18) as e [v 9vm /l 1 t(I / m) dF( y) 5 A9(m). Similarly, as in
*
t*
the decomposition in the text, t I 5 e0 (t(I / t 1 I) dt, but because it is
uncompensated taxable income that we deal with here, the integration takes place
at a constant income level. Using the Slutsky equation and rearranging leads to
t * I 5 e0t* I(1 2 t (I / E) 2 (t / 1 2 t) dt. Differentiating this with respect to m and
substituting into the first-order condition yields
t*

E5

v 9vm

]]
2]
l
m

t*

E S

0
t*

I
I
t
tI ] dt 1 ] 1 2 ]] dt
E
m
12t
0

tI
2 ]] ] dt dF( y) 5 A9(m).
1 2 t m
0

(A.1)

The last term on the left-hand side is the social benefit of a change in the elasticity.
The third term is the benefit, holding elasticity constant. The second term accounts
for a change in income, while the first is the direct effect on welfare. When the
elasticity is constant, the benefit of a change in elasticity may be expressed as
t*

tI
2 ]] ] dt 5 m
1 2 t m
0

t*

1E

t*

I
I dt 2 t I 2 tI ] dt .
E
*

(A.2)

The term in the brackets on the right-hand side is the deadweight loss of taxation,
corrected by potential revenue effects.

Appendix B. Proofs related to Section 3


Derivation of the compensated elasticity of taxable income: the standard
Slutsky equations for individual commodities are (Ci / t )u u 5 (Ci / t ) 2 (Ci /
E)I, where Ci / t u u is the compensated response of Ci , and E denotes exogenous
(non-taxable) income. Since, by assumption, the lump-sum tax T affects taxable
income, its effect on Ci is not the same as the effect of E, but they are linked
through the relation (Ci / T ) 5 2 t (Ci / E). Using these relations, (I / t )u u 5
2 em1 (Ci / t )u u di 5 I(1 2 g )t 21 , and the compensated elasticity of taxable
income is e 5 (t /I)(I / t )u u 5 1 2 g.

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

109

Lemma 1. The YitzhakiWilson (Y-W ) curve is upward sloping with dg / dT . 0


and dX / dT . 0.
The Y-W curve is defined by the budget constraint and Eq. (20). The budget line
may be used to define t(g, T ) 5 A(g ) 1 R 2 T /g (Y 2 T ). Differentiating this
expression and then simplifying yields t / T 5 2 1 2 tg /g (Y 2 T ) , 0 and t /
g 5 A9(g ) 2 t(Y 2 T ) /g (Y 2 T ) , 0. (To see that the numerator is negative,
observe that it is equal to t ln(t )(Y 2 T ) from Eq. (20)). Total differentiation of
Eq. (20) (with t 5 t(g, T )) yields dg / dT 5 2 (t 1 t ln(t ) 1 (Y 2 T ) ln(t )(t / T /
A0)(g ) 1 (Y 2 T ) ln(t )(t / g ) (it is negative because both the numerator and
denominator are positive). Finally, from the budget constraint, X 5 R 1 A(g ) 2 T,
so that dX / dT 5 A9(g ) dg / dT 2 1 , 0. h
Proposition 1. An increase in egalitarianism increases X and g.
Consider increasing the concavity of the welfare function, i.e., replacing w by
f(w), where f 9 . 0, f 0 , 0. The global optimum moves along a constant Y-W curve
(because neither the budget constraint nor Eq. (20) are affected) and by Lemma 1
either both X and g increase, while T falls (this is our thesis), or both X and g
decrease, while T increases. We can show that the second case is not possible.
Suppose that it was the case. Let v1 ( y) and v2 ( y) represent the level of utility of
person y with the optimal tax system under the initial and modified social welfare
function, respectively. There must be some individuals with v1 . v2 and some with
v2 . v1 , or else one of the these tax systems could never be optimal. The indirect
utility function is v1 ( y) 5 ( y 2 T )t g. Because we consider an increase in T, t g
must increase, or else v2 , v1 everywhere. Therefore, v2 ( y) 2 v1 ( y) is increasing
in y. Then, there must exist y* such that v2 ( y*) 5 v1 ( y*) and v2 ( y) , v1 ( y)y ,
y*. The same inequalities hold when we compare w(v1 ( y)) and w(v2 ( y)). The
concavity of f implies that f(w(v2 ( y))) , f(w(v1 ( y))) 1 f 9(w(v1 ( y)))[w(v2 ( y)) 2
w(v1 ( y))]. Integrating and rearranging this expression yields

E f(w(v ( y))) dF( y)


,E f(w(v ( y))) dF( y) 1 f 9(w(v ( y*))) E (w(v ( y)) 2 w(v ( y))) dF( y)
1 E [ f 9(w(v ( y))) 2 f 9(w(v ( y*))][w(v ( y)) 2 w(v ( y))] dF( y).
(B.1)
2

The second term is negative because e w(v1 ( y)) dF( y) is the maximized welfare
under the initial welfare function. Notice that for y , y*, we have v2 ( y) , v1 ( y)
and f 9(w(v1 ( y))) . f 9(w(v1 ( y*))) and both inequalities change their signs for y . y*
(this is implied by the concavity of f, and v1 increasing in y); thus, the third term
in negative. Thus, e f(w(v2 ( y))) dF( y) , e f(w(v1 ( y))) dF( y). This is a contradiction, however, because the left-hand side is the maximized welfare under the
modified social welfare function. Thus, the second case is not possible. h

110

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

Preliminaries for Lemmas 2 and 4: we proceed by considering the maximization of social welfare with respect to T, X and g (i.e., t is implicitly defined by
other variables). Define MCX(T, X(t, T, g );g ) ; (Y 2 T )21 e w9t g 21 ( y 2 T ) dF( y)
and MCT 1 (T, X(t, T, g ); g ) ; (1 2 tg )21 e w9t g dF( y) from the first-order
conditions (17) and (18). These are the marginal costs of raising a marginal dollar
of revenue by using distortionary and lump-sum taxes, respectively. When T is
used, collecting an additional dollar requires raising T by 1 / 1 2 tg dollars, and
reduces X by tg / 1 2 tg. Therefore, we introduce MCT(T, X(t, T, g );g ) ;
MCT 1 (1 2 tg ) 1 MCXtg, the marginal cost of raising revenue using T while
holding X constant. At the optimum, MCT 5 MCX 5 MCT 1 , by the first-order
conditions (17) and (18). Now consider a perturbation in the given value of g (a
change in any other parameter may be analyzed in the same way). Totally
differentiating the first-order condition MCT 5 MCX and the budget constraint
yields expressions for (X / g ) and (T / g ), as follows
MCT T 2 MCXT
X MCTg 2 MCXg
] 5 ]]]]] 1 A9(g ) ]]]]]
g
SOC
SOC

(B.2)

MCXX 2 MCT X
T MCXg 2 MCTg
] 5 ]]]]] 1 A9(g ) ]]]]]
,
g
SOC
SOC

(B.3)

where SOC (which, by the second-order condition, must be positive) equals


MCXX 2 MCXT 2 MCT X 1 MCT T . 0. These formulae decompose the total effect
of a change in g on X and T into substitution and revenue effects. The former
refers to the fact that a change in g alters the relative attractiveness of T and X as
ways to raise revenue; the latter refers to the fact that increasing g has a social
cost of A9, which has to be financed by T and X. In order to formally sign these
effects, it will prove useful to consider the ratio MCT /MCX, which equals
Ct 1 tg, where C ; e w9 dF( y) e ( y 2 T ) dF( y) / e( y 2 T )w9 dF( y) is a measure
of the distributional characteristic of the tax system.16 We will make use of the
following property of C :
Property B.1. A monotone, concave transformation of the social welfare function
increases C.
Proof. Let the original welfare function be w and the transformed one be g(w)
where g9 . 0, g0 , 0. Defining W(x) 5 e0x w9 dF( y) / e0` w9 dF( y) and G(x) 5 e0x
g9w9 dF( y) / e0` g9w9 dF( y), it suffices to compare e( y 2 T ) dW( y) and e( y 2 T )
dG( y). Obviously, lim y ` W( y) 5 1 and lim y ` G( y) 5 1. Concavity and monotonicity of g imply that g9( y 1 ) . g9( y 2 ) if and only if y 1 , y 2 . It follows that e0x
16

The inverse of C is the (normalized) Feldsteins (1972) measure of the distributional characteristic
of a tax.

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

111

g9w9 dF( y) . g9(x) e0x w9 dF( y) and ex` g9w9 dF( y) , g9(x) ex` w9 dF( y), so for
any x:
`

G(x) 5

E g9w9 dF( y)
1 1 ]]]]
E g9w9 dF( y)

x
x

21

E w9 dF( y)
1 1 ]]]]]
g9(x) E w9 dF( y)

21
.

g9(x)

x
x

21

5 W(x) .

This condition says that W(x) first-order stochastically dominates G(x) (although
the context is different). Thus e ( y 2 T ) dW( y) . e ( y 2 T ) dG( y), which directly
implies the desired result. h
Lemma 2. Consider MCT /MCX. For given X and T, the transformation of the
social welfare function affects this ratio only through its effect on C. Property B.1
guarantees that C increases when the social welfare function becomes more
egalitarian, thus increasing the relative marginal cost of T versus X, i.e, (MCT /
MCX) / C . 0. Therefore, MCTC MCX 2 MCXC MCT . 0. At the optimum,
MCX 5 MCT . 0, so we must have MCXC , MCTC . Analogously to the derivation of Eqs. (B.2) and (B.3), this implies a substitution response from T to X
(because SOC . 0). It is the only effect, because the transformation of the social
welfare function does not have a revenue aspect, concluding the proof. h
Lemma 4. Consider term MCT /MCX, defined as a function of X, T and g. When g
changes t must adjust to keep X 5 tg (Y 2 T ) constant. Consequently, dt / dg 5 2 t /
g , 0. Under the isoelastic social welfare function, C / g 5 C / t 5 0, thus
(MCT /MCX) / g 5 C t /g . 0. Because at the optimum MCX 5 MCT . 0, this
implies that MCXg , MCTg . Now recall Eqs. (B.2) and (B.3), in which the first
term in both formulae is a substitution effect and the second one is a revenue
effect. We have just signed the numerators of the substitution terms so, given that
SOC . 0, the substitution response is to increase X and decrease T. When A9(g ) is
small enough, it must dominate the revenue effect. h
Proposition 2. The previous optimum (g *, X*) still lies on the original Mirrlees
curve. As a result, there are only second-order changes in the solutions of the
Mirrlees problem in a neighborhood of the optimum. The changing administrative
cost does, though, affect the Y-W curve: it changes the optimal mix of t and g to
raise a given amount of distortionary tax. In fact, the Y-W curve must shift
upward, so that for a given g, the optimal X is higher. To see this, consider if it
were otherwise, so that the optimal X declines which implies, via the budget
constraint, that the corresponding value of T increases. Recall that X 5 tg (Y 2 T )
so that, even though an increase in T affects X, t must fall to keep gross revenue

112

J. Slemrod, W. Kopczuk / Journal of Public Economics 84 (2002) 91 112

(T 1 X) constant. But the right-hand side of expression (20), which has to be


satisfied on the Y-W curve, is increasing in t, and so an increase in T and a decline
in t imply that the right-hand side of this expression needs to fall, as well. Yet, by
assumption, the left-hand side of (20) increases, leading to a contradiction. h

References
Atkinson, A.B., Stiglitz, J.E., 1980. In: Lectures on Public Economics. McGraw-Hill, London, New
York.
Auten, G., Carroll, R., 1999. The effect of income taxes on household behavior. Review of Economics
and Statistics 81 (4), 681693.
Carroll, R., 1998. Do taxpayers really respond to changes in tax rates? Evidence from the 1993 Act.
Office of Tax Analysis Working Paper 78, US Department of Treasury.
Cremer, H., Gahvari, F., 1993. Tax evasion and optimal commodity taxation. Journal of Public
Economics 50 (2), 261275.
Cremer, H., Gahvari, F., 1994. Tax evasion, concealment and the optimal linear income tax.
Scandinavian Journal of Economics 96 (2), 219239.
Cremer, H., Gahvari, F., 1996. Tax evasion and the optimum general income tax. Journal of Public
Economics 60 (2), 235249.
Feldstein, M.S., 1972. Distributional equity and the optimal structure of public prices. American
Economic Review 62 (1 / 2), 3236.
Feldstein, M.S., 1995. The effect of marginal tax rates on taxable income: A panel study of the 1986
Tax Reform Act. Journal of Political Economy 103 (3), 551572.
Feldstein, M.S., 1999. Tax avoidance and the deadweight loss of the income tax. Review of Economics
and Statistics 4 (81), 674680.
Kaplow, L., 1990. Optimal taxation with costly enforcement and evasion. Journal of Public Economics
43 (2), 221236.
Lucas, R.E., 1976. Econometric policy evaluation: A critique. Journal of Monetary Economics 1 (2),
1946.
Mayshar, J., 1991. Taxation with costly administration. Scandinavian Journal of Economics 93 (1),
7588.
Moffitt, R., Wilhelm, M., 2000. In: Slemrod, J., (Ed.) Does Atlas Shrug? The Economic Consequences
of taxing the rich, Russell Sage Foundation and Harvard University Press, NY and Cambridge.
Okun, A.M., 1975. In: Equality and Efficiency: The Big Tradeoff. The Brookings Institution,
Washington, DC.
Slemrod, J., 1994. Fixing the leak in Okuns bucket. Optimal progressivity when avoidance can be
controlled. Journal of Public Economics 55 (1), 4151.
Slemrod, J., 1998. Methodological issues in measuring and interpreting taxable income elasticities.
National Tax Journal 51 (4), 773788.
Slemrod, J., 2001. A general model of the behavioral response to taxation. International Tax and Public
Finance 8 (2), 119128.
Wilson, J.D., 1989. On the optimal tax base for commodity taxation. American Economic Review 79
(5), 11961206.
Yitzhaki, S., 1979. A note on optimal taxation and administrative costs. American Economic Review
69 (3), 475480.

Das könnte Ihnen auch gefallen